Stable Diffusion: image generation goes open

In one sentence Stability AI publicly releases weights and code of a text-to-image latent diffusion model that runs on a consumer GPU. AI image generation leaves the cloud.

Verified Official source

ShareLinkedIn X

Stability AI releases Stable Diffusion: a model that, given a text prompt ("an astronaut on horseback, Renaissance style"), produces an image in seconds.

The difference from DALL·E 2, released a few months earlier, is that here the model weights are public. Anyone with a gaming GPU can download it and run it locally, sending nothing to an external server.

From that moment everything explodes: desktop UIs, Photoshop plugins, forks, fine-tuning on personal styles. "Generative AI" stops being a closed cloud service.